Towards User-Centric Rate Adaptations for VoIP Traffic

Voice traffic by nature is high in data volume and sensitive to network impairments. Bandwidth requirement for voice calls is, therefore, higher. An increasingly serious dilemma, as the amount of VoIP traffic increases, is that it will not be fair to give priority to the multimedia traffic and starve its best-effort counter part; yet the voice data delivered might not be perceptible if each voice call is limited to the rate of an average TCP flow. We propose to approach this problem from a user-centric view and adapt the sending rate of the voice calls based on the user satisfaction. Such a user-friendly rate adaptation mechanism will also be congestion-friendly, although not strictly TCP-friendly [1]. There has not been a promising rate adaptation mechanism yet for userand congestion-friendly VoIP services. In [2], the authors propose a TCP-like rate adaptation mechanism. In that, the sending rate is adjusted in an additive-increase-multiplicative-decrease fashion based on the measured loss rate and delay. [3] [4] propose to use speech quality indicator such as E-Model as the metrics to adapt the sending rate. However, the number of parameters required to calculate the E-Model value is high, twenty-one to be exact. Some parameters such as mouthto-ear delay are not easily accessible. Many of the parameters are over-simplified in the existing works. Exploiting the User Satisfaction Index (USI) derived from a recent study on Skype traffic [5], we propose a rate adaptation mechanism for Skype voice call which requires only 3 parameters that can be measured and computed in real time. Although the USI is specific to the voice codec used by Skype and might not be generally applicable for all VoIP calls, our work serves as a necessary next step towards user-centric rate adaptation for VoIP traffic on the Internet. The preliminary experimental results show that rate adaptation mechanisms based on USI may result in more satisfying voice communication experience. *